Room-temperature bonding of GaN and diamond via a SiC layer

Abstract A GaN-on-diamond structure is the most promising candidate for improving the heat dissipation efficiency of GaN-based power devices. Room-temperature bonding of GaN and diamond is an efficient technique for fabricating this structure. However, it is extremely difficult to polish diamond to an average roughness (Ra) below 0.4 nm, especially for polycrystalline diamond. In this work, Room-temperature bonding of GaN and rough-surfaced diamond with a SiC layer was successfully achieved by a surface-activated bonding (SAB) method. The diamond surface’s initial Ra value was 0.768 nm, but after deposition of the SiC layer, the Ra decreased to 0.365 nm. The SiC layer formed at the as-bonded GaN/diamond interface was amorphous, with a thickness of about 7 nm. After annealing at 1000-°C, the amorphous SiC layer became polycrystalline, and its thickness increased to approximately 12 nm. These results indicate that the deposition of a SiC layer on diamond can efficiently lower the diamond surface’s roughness and thus facilitate room-temperature bonding.


Introduction
Devices based on GaN have been extensively studied for high-frequency and high-power applications. The output power density of GaN-based devices is limited by self-heating during their operation. Heat accumulation increases the device's temperature, which degrades its output characteristics. As the power density increases, the self-heating effect becomes more obvious. Hence, to increase the output power density of these devices, we will need to solve the problem of heat accumulation in the channel region. High-power GaN-based devices have normally been fabricated on SiC substrates [1]. Although the thermal conductivity of SiC is higher than that of Si or sapphire, it is still not high enough to fully meet the heat dissipation requirements of GaN-based devices [2][3][4][5][6].
In contrast, diamond has a thermal conductivity several times higher than that of SiC, and it is being extensively studied as a heat-spreading substrate for GaN-based devices [4,[7][8][9][10]. In particular, GaN-ondiamond structures have attracted more attention because they can improve the cooling efficiency of devices [11][12][13]. There are two main ways to form such a structure. The first is to remove Si from the GaN-on-Si substrate and deposit diamond on the exposed GaN surface by using a buffer layer such as SiN x or AlN [14,15]. The second way is to remove Si from the GaN-on-Si substrate and bond it to a diamond substrate by using a metal or amorphous Si interlayer [16][17][18][19]. In the first approach, the buffer layer contributes a large thermal resistance and causes low thermal conductivity in the deposited diamond because of low crystal quality, which prevents the desired thermal management from being achieved. In the approach, the interlayer increases the thermal boundary resistance (TBR) between the GaN and diamond, and the low electrical resistivity of the interlayer produces parasitic capacitance when a device works at high frequency, which is not suitable for high-frequency devices.
On the other hand, low-temperature wafer direct bonding technology is a promising approach for the integration of GaN and diamond, as it can overcome the lattice and thermal expansion mismatches between the bonded materials. Specifically, direct bonding at low temperatures has been demonstrated for Si/diamond, Ga 2 O 3 /diamond, InGaP/diamond, and GaAs/diamond [20][21][22][23]. Although these bonding processes were conducted at low temperatures, the bonded interfaces revealed excellent thermal stability through control of the interface structure. We previously demonstrated direct bonding of GaN and diamond by a surface-activated bonding (SAB) method at room temperature [24]. The bonded GaN/diamond interfaces exhibited thermal stability as high as 1000-°C and excellent practicality for GaN-based devices. However, wafer direct bonding technology requires the bonded materials to have very high surface flatness, especially as material's mechanical strength increases. To achieve direct bonding of diamond and dissimilar materials, the average surface roughness (Ra) is usually required to be less than 0.4 nm [24][25][26][27]. Diamond has an extremely high hardness that makes it difficult to polish to an atomically flat surface, especially for polycrystalline diamond because of the different grain sizes and orientations. Hence, it is necessary to develop a wafer-bonding technology that can bond diamond even with poor surface flatness.
In this work, we explored room-temperature bonding of diamond and GaN with a SiC layer by the SAB method, and we evaluated the thermal stability of the bonded GaN/diamond interface at 1000-°C in N 2 gas at ambient pressure. The interface's structure and atomic behavior were systematically investigated by transmission electron microscopy (TEM) and energy-dispersive X-ray spectroscopy (EDX). The chemical bonding states of the carbon atoms at the interface were also investigated by electron energy-loss spectroscopy (EELS).

Experimental
We performed roo-temperature bonding experiments with substrates made by growing GaN epitaxial layers on Si (111) and diamond synthesized at high temperature and high pressure (HTHP). The substrate size was 3 mm × 3 mm × 0.36 mm. We also used GaN epitaxial substrates comprising a 1000-nm-thick GaN layer, and a 100-nm-thick AlN buffer layer grown on a Si (100) substrate by metal-organic chemical vapor deposition (MOCVD). Before bonding, the diamond surface was mechanically polished to obtain a good flat surface. A SiC layer with a thickness of about 15 nm was deposited on the diamond surface by RF sputtering after taking the SiC etching rate of Ar fast atom beam irradiation during the bonding process into account. The deposited SiC layer thickness was measured by a step meter. Figure  1 The diamond substrates with SiC were cleaned with a mixture of sulfuric acid and hydrogen peroxide (H 2 SO 4 :H 2 O 2 = 4:1) at 80-°C for 10 min, rinsed with deionized water for 3 min, and dried by an N 2 flow. The GaN epitaxial substrates were cleaned with acetone and isopropyl alcohol in an ultrasonic bath for 300 s and then dried by an N 2 flow. Next, the substrates were placed in an SAB facility for room-temperature bonding. More details of the bonding procedures are described in our previous work [24]. After bonding, the Si substrates used for the GaN growth were removed by mechanical polishing and chemical wet etching.
The crystal structure and chemical composition of the GaN/diamond interface before and after annealing at 1000 °C for 1 min were investigated by TEM and EDX under scanning transmission electron microscopy (STEM) with a JEOL JEM-2200FS analytical microscope. The bonding states of the carbon atoms near the GaN/diamond interface were also investigated by EELS under STEM. Carbon K-shell edge spectra were taken between 280 and 380 eV at an acceleration voltage of 200 kV. The TEM samples were fabricated by the focused ion beam (FIB) technique (Helios NanoLab600i; Thermo Fisher Scientific) at room temperature.

Results and discussion
The optical microscopy images of the as-bonded GaN/ diamond sample before and after removing the Si substrate are shown in Figure 2 (a)and (b), respectively. An about 85% bonded area was obtained. The unbonded area was due to a low uniformity of the diamond substrate thickness. After removing Si, a reduction in the bonded area of about 3% was observed. The bonding interface pealing occurs near the non-bonded area, which is attributed to the weak bonding strength. These results indicate that an entire area bonding of GaN and diamond can be achieved using a high uniformity diamond substrate in the thickness. diamond substrate, which corresponds to the diamond ( ) 220 plane. The lattice fringe thickness was estimated to be about 3.7 nm. Application of a fast Fourier transform (FFT) to the intermediate layer revealed that it was amorphous. Thus, the intermediate layer formed at the as-bonded interface consisted of two parts: an amorphous layer, and a crystal-defect diamond layer. The thickness of the amorphous layer was determined to be about 7.0 nm. Figure 4(a-c) show EDS mappings and X-ray intensity profiles of the as-bonded GaN/diamond interface for N, Ga, Fe, O, C, Ar, and Si atoms (highlighted in olive green, green, blue, red, cyan, black, orange, respectively), and enlarged X-ray intensity profiles for Fe, O, and Ar atoms. We observed an atomic layer of O and Ar and two atomic layers of Fe, which were due to contaminants originating from the vacuum chamber during the bonding and SiC film deposition processes. In the Fe atom intensity profile, the peak adjacent to the GaN substrate was higher than the peak adjacent to the diamond substrate because of the larger energy used in the bonding process. Previous research has shown that the formation of Fe atoms on the surface is only activated by Ar beam irradiation during the bonding process, which explains why the peak in the Fe atom intensity profile was located at the bonding interface [28]. As expected, the peak in the Ar atom intensity profile was at the deposition interface because of the Ar sputtering gas used in the SiC film deposition process. The distance between the bonding and deposition interfaces was determined to be about 9 nm, which agreed  with the thickness of the amorphous layer observed at the interface. We can see that Si atoms diffused into the GaN substrate adjacent to the bonding interface, and Ga and N atoms diffused into the amorphous layer. Figures 5(a-c) show an EELS spectra line scanned across the as-bonded GaN/diamond interface with a step of 0.2 nm, EELS spectra of amorphous carbon, diamond, and amorphous SiC, and sp 2 ratios obtained from the EELS spectra. π* and σ* peaks were observed in the EELS spectra at 285 and 292 eV, respectively. The respective peaks represent the sp 2 trigonal coordination and sp 3 tetrahedral coordination of carbon in graphite or amorphous carbon and diamond. The sp 2 ratios shown in Figure 5(c) were calculated by integrating the areas of the π* and σ* peaks [24]. Because the carbon bond of SiC is only a σ bond, the K-shell (1 s) electron of carbon is excited to the σ* orbital, and only the σ* peak should be observed in the EELS spectrum. The fact that the π* peak was also observed indicates that a π bond like that of amorphous carbon existed in the amorphous SiC. The sp 2 ratio of the intermediate layer increased in the range of 0-7 nm and reached about 96% at 7 nm, after which it remained essentially the same, in the range of 7-15 nm. The sp 2 ratio of the amorphous SiC was calculated to be about 96%. Thus, we conclude that the main components in the range of 0-7 nm were amorphous carbon and diamond, while the main component in the range of 7-15 nm was amorphous SiC. The thicknesses were measured to be about 7 and 8 nm, respectively. The distribution range of amorphous SiC also agreed with the thickness of the amorphous layer formed at the as-bonded GaN/diamond interface.
Next, for the case of annealing at 1000 °C, Figure  6 For the GaN/diamond interface annealed at 1000 °C, Figure 7(a-c) shows EDS mappings and X-ray intensity Lastly for the annealed interface, Figure 8(a-d) shows an EELS spectra line scanned with a step of 0.2 nm, EELS spectra of diamond and polycrystalline SiC, and sp 2 /sp 3 and sp 2 ratios obtained from the EELS spectra as a function of the scanned distance. A π* peak was again observed in the EELS spectrum of polycrystalline SiC, which indicates that it also had the π bond. The sp 2 /sp 3 ratio of the polycrystalline SiC was determined to be about 4.6%. The sp 2 /sp 3 ratio in the intermediate layer in the range of 7-18 nm was also about 4.6%, which means that the intermediate layer was poly crystallin SiC. The thickness of the part composed of amorphous carbon and diamond greatly decreased to about 3 nm. The actual thickness would be even smaller if no damage were caused by the FIB during the TEM sample fabrication process.
For the amorphous SiC layer formed at the as-bonded interface, we observed a large discrepancy between the thickness of the deposited SiC on diamond and the observed SiC at the interface. This discrepancy was due to the SiC being etched by the Ar beam irradiation during the bonding process. A   similar result was reported for SiC etched by Ar plasma irradiation [29]. The defect-crystal diamond layer composed of amorphous carbon and diamond was induced by the Ar plasma irradiation during the deposition of the SiC layer. Defect-crystal diamond formed by Ar irradiation was previously reported at a GaAs/ diamond heterointerface [22]. After annealing at 1000 °C, the thickness of the defect-crystal diamond layer decreased, while that of the intermediate layer increased. There are two possible explanations for this discrepancy. The first is that the carbon atoms in the defect-crystal diamond layer could have reacted with silicon atoms to form SiC 30. The second is that the amorphous carbon in the defect-crystal diamond could have been converted to diamond, while the intermediate layer's thickness was increased by meltback etching of Si by Ga. Regarding the latter explanation, there have been reports of amorphous carbon in defect-crystal diamond being converted to diamond at a GaN/diamond interface annealed at 1000 °C [24], and of melt-back etching of Si occurring at a GaN/Si interface annealed at 1050 °C 31. In addition, we observed a destroyed GaN layer at the annealed interface. Thus the second explanation seems more reasonable. However, the first possibility cannot be completely dismissed, and it is possible that both mechanisms occurred simultaneously at the interface.
It is also important to note that the diamond surface's average roughness was significantly improved after deposition of the SiC layer. We were previously unsuccessful in attempting to bond a rough diamond surface to GaN without depositing a SiC layer. Here, the success in bonding GaN and diamond with the SiC layer indicates that SiC deposition will become an important technique to solve the problem of poly crystalline diamond surface polishing. Moreover, the reported thermal resistance of a GaN/SiC interface [32] and the theoretical thermal resistance of a SiC/diamond interface [33] are very low, and these are very useful characteristics for improving the heat dissipation efficiency of power devices.

Conclusion
We achieved room-temperature bonding of GaN and rough-surfaced diamond by using a SiC layer with the SAB technique. After the deposition of the SiC layer, the diamond surface's Ra value was greatly lowered. The SiC layer was found to be an amorphous layer with a thickness of about 7 nm. After annealing at 1000 °C, the SiC layer thickness slightly increased because of silicon and carbon atoms reacting to form SiC. In addition, the amorphous SiC layer became polycrystalline after the annealing process. No voids were observed at the bonding interface even after annealing, which indicates that the interface with the SiC layer had excellent thermal stability. These results show that the deposition of a SiC layer can reduce the roughness of a diamond surface and facilitate room-temperature bonding of polycrystalline diamond and dissimilar materials.

Disclosure statement
No potential conflict of interest was reported by the authors.